hide
Free keywords:
-
Abstract:
We are currently developing a semi-automatic system for the reconstruction of 3-D human movement data from complex natural scenes, such as movie clips. As part of this project we have collected a database of walking patterns from twenty well-known male actors. The goal of this initial study was to assess whether isolated 2-D motion cues (point-light walkers created via manual tagging) could provide sufficient information for the recognition of famous gaits. Previous research has indicated that familiar individuals (eg work colleagues) can be recognised as point-light figures. Does memory for famous individuals also include characteristic movement patterns? Observers were shown point-light animations depicting several step-cycles from different actors, filmed from approximately 3/4 view. As the animations were extracted from commercial footage, exact camera position, gait cycle (eg number of repeated steps), and extraneous behaviours (eg additional hand movements) could not be controlled. Animations were, however, approximately normalised for size, and the global translation cues were removed. Using both a direct 6-alternative face-to-gait matching test and a standard 2-alternative forced-choice task, we found levels of performance that did not differ from chance. Item analysis revealed that neither self-reported familiarity with the actors nor confidence ratings provided accurate predictors of performance.
